# 1. Foundational concepts of prompt engineering

Before we jump into product management use cases, let's ensure we understand what prompt engineering actually means and why it matters.

Prompt engineering is the practice of designing and refining prompts — questions or instructions — to elicit specific and useful responses from AI models. In essence, a prompt is how you "ask" the AI to do something, and prompt engineering is about asking in the optimal way. Think of it as the interface between your human intent and the machine's output.

At its core, prompt engineering involves a few key ideas:

- **Clarity and context:** The way you phrase a request greatly influences an AI's understanding. A vague prompt leads to an aimless answer, whereas a clear, detailed prompt guides the AI toward the information you want. For example, simply asking "Tell me something" may yield virtually anything, but asking "Can you summarize today's stock market status?" gives a focused, relevant response. Always provide enough context so the AI isn't guessing what you need.
- **Specific instructions:** The more precise you are about the desired outcome, format, or style, the better. Just as you'd give a junior PM clear guidance, you should tell the AI exactly what you want. For instance, instead of "Write a product description," you might say "Write a 3-5 sentence product description for our mobile app, in an upbeat tone, highlighting the key benefit to users." Being specific about length, tone, and content will yield a more useful result.
- **Role prompting:** This is my favorite. You can ask the AI to take on a role or perspective. This is like setting context. If you begin a prompt with "You are an expert fintech product manager…", the AI will tailor its tone and knowledge accordingly. Assigning a role instantly shapes the expertise and style of the answer. For example, compare the below prompts:
  - *Ineffective prompt:* "Explain how to improve our onboarding."
  - *Effective prompt:* "As a customer success manager, explain how we can improve the onboarding experience of our SaaS product to reduce drop-off, providing 3 concrete suggestions."

  In the effective prompt, we gave the AI a persona (customer success manager) and a specific task with detail – likely yielding a more on-point answer.
- **Few-shot examples:** Sometimes showing one or two examples of the kind of output you expect can guide the model. This is known as few-shot prompting. If zero-shot (no examples) works, great – but if not, you can include a sample input-output pair in your prompt. For instance, when extracting data, you might say: "Convert the text to key-value pairs. Example: 'Product: X, Price: $10' -> {Product: X, Price: 10}. Now do the same for this text: [your text]." This gives the AI a template to follow. Use this technique sparingly, as adding examples consumes the prompt length, but it can improve accuracy if the model was initially confused.
- **Understanding AI's limits:** Remember that AI models don't truly think or know things outside their training data. They generate best-guess completions based on patterns. They also have a limited "memory" (context window) of the conversation. For tools like ChatGPT, this might be a few thousand tokens (roughly several pages of text). Keeping prompts focused ensures you don't exceed these limits. Also, if you ask for something extremely broad or multifaceted, the model might miss some aspects. We need to sometimes break complex requests into smaller prompts – we'll explore this shortly as an advanced technique.

Why does all this matter for you as a product manager? Because a well-crafted prompt can be the difference between an AI assistant that's actually useful and one that wastes your time. Prompt engineering ensures the AI understands the nuances and intent behind your query. Think of early voice assistants – if you vaguely say "play something", you get random results, but if you say "play a relaxing jazz playlist on Spotify", you get exactly what you wanted. With generative AI, the same principle applies, but on a broader range of tasks.

Finally, keep in mind that prompt engineering is part art and part science. There's no single "perfect prompt" – it often involves iterating and refining. It's similar to refining product requirements: rarely perfect on the first draft. In fact, prompt engineering has evolved through practice. Initially, many interactions were simple Q&As, but users learned to add structure, context, and personality to prompts to get better results. It's an emerging skill set, and as such, it improves with experimentation and practice. Don't worry – by the end of this guide, you'll have plenty of examples and frameworks to kick-start your own prompt engineering practice.